# Claude Context — Claudify Content Specialist

This project uses Claudify Content Specialist — the complete operating system for content production, brand voice, and audience growth.
Always read `.claude/memory.md` before taking action.

## Quick Start
- Run `/start` to begin your session
- Run `/content {brand}` to run the full content production pipeline (script → asset → variants)
- Run `/post {platform} {topic}` to draft a single post (X, LinkedIn, IG, Threads)
- Run `/thread {topic}` to draft an X / Twitter thread
- Run `/blog {keyword}` to draft a long-form blog post (SEO-aware)
- Run `/newsletter {action} [N]` for the full newsletter pipeline (strategy → draft → architect → build → audit → render-test → preview → send → analyse)
- Run `/launch-post {platform}` to draft a Product Hunt / Hacker News / Indie Hackers launch
- Run `/readme {project}` to draft a high-conversion README + docs
- Run `/feedback` fires automatically after content verdicts to log learnings
- Run `/sync` mid-session to refresh memory
- Run `/wrap-up` at end of day
- Run `/audit` to verify recent work quality
- Run `/safe-clear` to flush context and resume fresh

## Key Files
- Memory: `.claude/memory.md` (read this for current context)
- Knowledge Base: `.claude/knowledge-base.md` (content rules — mandatory constraints, read before every task)
- Task Board: `Task Board.md`
- Scratchpad: `Scratchpad.md` (quick capture, processed during /sync)
- Daily Notes: `Daily Notes/` (created automatically by /start)
- Knowledge Nominations: `.claude/knowledge-nominations.md` (candidate learnings — auditor reviews)
- Command Index: `.claude/command-index.md` (all commands with triggers and tools)
- Content Engine Memory: `.claude/agent-memory/content-creator/MEMORY.md` (production log, calibration, approved patterns)
- Brand Memories: `.claude/agent-memory/content-creator/brand-memories/{brand}.md` (per-brand voice, ICP, history)
- Newsletter System: `.claude/agent-memory/newsletter-orchestrator/MEMORY.md` (issue history, voice patterns, render-pipeline learnings)

## System Architecture
- **Agents** (`.claude/agents/`): 8 specialist subagents with persistent memory
  - **Content production (4):**
    - `auditor` — Quality gate. Reviews all content output. Promotes confirmed learnings to knowledge base.
    - `content-creator` — Lead production agent. 6-mode pipeline across all platforms. Reads brand memory + voice rules before drafting.
    - `content-strategist` — Editorial planner. Owns topic backlog, type rotation, content calendar. Writes executable briefs.
    - `brand-voice-keeper` — Voice consistency across sessions. Manages brand-memory architecture. Catches drift before it ships.
  - **Newsletter pipeline (4):**
    - `newsletter-orchestrator` — Coordinates the 9-phase newsletter pipeline.
    - `drafter` — Writes issue body content from a strategist brief.
    - `newsletter-auditor` — 25 mechanical + 13 semantic + regression checks.
    - `stylist` — Visual rendering gate. 20 structural checks + Playwright at 4 viewports.
    - `analyst` — Post-send learning. Day-of-send + day-7 reports.
    - `prompt-architect` — Designs the god-tier prompt. Tests against 3 reference codebases.
- **Commands** (`.claude/commands/`): 13 commands (5 system + 8 content)
- **Hooks** (`.claude/hooks/`): 14 hooks (9 base + 5 newsletter-specific)
- **Newsletter system** (`newsletter/`): Full production system (if installed)

## Memory Architecture (6 Tiers)
1. **memory.md** — Active session context (current brands, priorities, open tasks)
2. **Content Creator Memory** (`.claude/agent-memory/content-creator/MEMORY.md`) — Production log, calibration data, approved voice patterns, rejected patterns
3. **Brand Memories** (`.claude/agent-memory/content-creator/brand-memories/{brand}.md`) — Per-brand voice samples, ICP, signature phrases, banned phrases, content history
4. **Knowledge Base** (`.claude/knowledge-base.md`) — System-wide content rules (auditor-gated, pre-seeded with hard rules)
5. **Knowledge Nominations** (`.claude/knowledge-nominations.md`) — Candidate learnings pipeline awaiting auditor promotion
6. **Daily Notes** — Chronological session history and handoff records

## Command Awareness

All agents can invoke system commands. Read `.claude/command-index.md` for the full catalog.

- **Self-execute**: If you have the tools a command requires, read `.claude/commands/{name}.md` and follow the procedure directly.
- **Recommend**: If you lack the tools, output `RECOMMEND: /command [args] — [reason]` for the orchestrator.
- Agents should proactively invoke commands when trigger conditions match.

### Content Feedback Auto-Detection (mandatory)

After ANY content production output, the user's next substantive response is treated as a verdict:
- **Approve signals**: "approved", "looks good", "ship it", "go", "yes", "love it", "perfect", "post it"
- **Revise signals**: "change X", "too Y", "needs Z", "not quite", "almost", specific corrections
- **Reject signals**: "no", "scrap it", "start over", "rejected", "not right"

When detected: self-execute `/feedback` procedure immediately. Never ask "should I log this?" — just do it.

## Retrieval Map — Where to look for what

| You need... | Check first | Then |
|---|---|---|
| What am I working on now? | `memory.md` → Now | Task Board → Today |
| Brand voice for client X | `agent-memory/content-creator/brand-memories/{brand}.md` | brand-voice-keeper |
| Voice rules and constraints | `knowledge-base.md` → Hard Rules — Voice & Tone | brand-voice-keeper |
| Approved voice patterns (3+ confirmations) | `agent-memory/content-creator/MEMORY.md` | brand-memory file |
| Rejected patterns to avoid | `agent-memory/content-creator/MEMORY.md` → Watch list | brand-memory file |
| What commands exist | `.claude/command-index.md` | `.claude/commands/{name}.md` |
| What went wrong before | `knowledge-base.md` → Known Failure Modes | content-creator MEMORY |

## Agents — When to Invoke Each

| Task | Agent |
|---|---|
| Production pipeline (any platform/format) | `content-creator` |
| Topic selection, content calendar, brief production | `content-strategist` |
| Voice drift / consistency check before publish | `brand-voice-keeper` |
| Newsletter (any phase: strategy → analyse) | `newsletter-orchestrator` |
| Quality review of any content output | `auditor` |
| Post-publish learning capture | `auditor` (via `/feedback`) |

## Context Health

**Automatic safety net (hooks):**
- `PreCompact` hook saves state before auto-compaction
- `SessionStart(compact)` hook restores context after compaction
- `SessionStart(user)` hook resets stale gate files on every fresh session

**Self-monitoring:**
- After ~30+ tool calls or 3+ large file reads: run `/safe-clear` proactively
- If a "compacting conversation" warning appears: run `/safe-clear` immediately
- When completing a full content piece: run `/safe-clear` before next unrelated piece
- When switching between brands: acknowledge the boundary, prefer `/safe-clear`

## Content Operating Principles

**Voice is calibrated, never improvised.** Every brand has a `brand-memory.md` with 3+ calibration paragraph pairs. Style guides written abstractly fail; concrete examples lock voice across sessions.

**The audit is non-negotiable.** Every content piece runs through the auditor in fresh context before delivery. Skipping the audit cycle has caused every shipped-broken piece to date.

**Specific over generic, always.** Numbers beat magnitudes. Named tools beat "your tool of choice". Concrete moments beat abstract preambles.

**Brand-positioning is load-bearing.** The brand never appears as the entity with the problem the brand sells the solution to. Frame from the outside: pattern observation, authority position.

**Distribute or it didn't ship.** Every long-form piece generates 3-5 short-form derivatives. Without distribution, the piece reaches only the on-platform audience.

**Track everything.** Every shipped piece logs to memory. Memory grows monotonically; the system gets sharper as pieces accumulate.

**Mobile-first.** 60%+ of email opens are on mobile. 50%+ of social impressions. Render at 320px width before considering shipped.

**Anti-slop discipline.** AI-detector tells get banned at the hook level. Audience sophistication has eclipsed ChatGPT-default voice; sounding AI-generated kills credibility within seconds.

## Maintenance
- Keep memory.md compact (<100 lines)
- Keep brand-memory files curated (archive content history older than 90 days)
- Auditor proposes voice/format rule revisions — user approves before changes apply
- Review incident log during /sync and /wrap-up
- Run quarterly review per brand: confirm voice patterns, retire stale ones